Context Aggregation and Analysis: A Tool for User- Generated Video Verification
By Olga Papadopoulou, Dimitrios Giomelakis, Lazaros Apostolidis, Symeon Papadopoulos, Yiannis Kompatsiaris of CERTH-ITI. Dec 2019
Context Aggregation and Analysis: A Tool for User- Generated Video Verification. By Olga Papadopoulou et al
1. Context Aggregation and Analysis: A Tool for User-
Generated Video Verification
Olga Papadopoulou, Dimitrios Giomelakis, Lazaros Apostolidis,
Symeon Papadopoulos, Yiannis Kompatsiaris
Journalism department of AUTH
12 December 2019
2. CERTH was established in 2000
• 5 institutes, >700 employees (ITI is the largest with >300 employees)
• 1200 projects, 1100 international collaborations
• Among top-10 EU institutions in attracting competitive research projects
MKLab is among the biggest ITI labs with >60 researchers (20+ post-docs)
• Key areas: multimedia, social media, computer vision, data mining, machine
learning
• Since 2003, involved in more than 60 research projects and published >600
research papers
MeVer is a team that develops technologies and services for the detection of media-
based disinformation - https://mever.iti.gr/web/
• Datasets - https://mever.iti.gr/web/resources/
• Services - https://mever.iti.gr/web/resources/
• Context Aggregation and Analysis
• Near duplicate detection
• Image forensics
Follow us https://twitter.com/meverteam
MeVer @ MKLab - CERTH-ITI
3. InVID – WeVerify Plugin
https://weverify.eu/verification-plugin/
>17.000 users all over the world
CAA is integrated as component of the Verification plugin (Analysis)
4. User generated videos
Staged
Reuse
Tampered
Croatia right now - Fifa World
Cup Final 2018
$250,000 car gets windshield SMASHED
by kid on a skateboard!!!
https://www.youtube.com/watch?v=VBa4D9D6Gng
Lion Takes Revenge On Trophy Hunter!
https://www.youtube.com/watch?v=l7yt-VPYtOA
https://www.youtube.com/watch?v=l7yt-VPYtOA
5. Context Aggregation and Analysis
Platform APIs
CAA
COMPONENTS
Verification
report
https://caa.iti.gr
A tool that aims to facilitate the verification of user-generated
videos.
Provide URL Start Verification
6. Video and account metadata
Metadata about the video and the source
that posted the video:
• Video title
• Video description
• Create time
• …..
• Mentioned locations – extracted
by the text (title, description)
• Channel name
• Channel creation time
7. Video Comments
1. All comments left below the video
2. Verification comments – filtered by a list of
predefined verification related keywords
helpful for verification
3. Link comments, comments that contain
links to external sources
4. Free text comments, the user can provide
his/her own keyword and create subset
8. Reverse image search
The video thumbnails as returned by the
Platform APIs
Buttons for applying reverse image search
are included below each thumbnail for:
1. Google reverse image search
2. Yandex reverse image search
9. Twitter timeline
• Tweets sharing the URL of the video in
question are collected and visualized in
a timeline.
• The red line indicates the time that the
video was posted.
• Clicking on each box (tweet) the text of
the tweet appears along with a link to
the Tweet Verification Assistant ‘check
tweet veracity’ which extracts a score
indicating the tweet credibility.
10. Verification AI score
A score in the range of [0 1].
The higher the score the less credible the
video is.
The score is extracted using a machine
learning method. Although it is helpful,
indicator the accuracy of the algorithm is
~70% so it should be considered for the
verification process but the user should
not leverage only on it for the final result.
12. Twitter Timeline
Twitter timeline:
The tweets sharing the submitted video URL for YouTube and Facebook
videos.
The retweets of a submitted Twitter video.
A tweet is posted couple of hours
after the Video was shared on
YouTube (redline) and explains that
the claim of ISIS being the
target of the bombing is false.
Claim: Bombing over ISIS area
13. User Study
Tasks:
Debunking the 200 fake videos of the FVC
Verifying the 180 real videos of the FVC
Users:
A male with journalistic background
A female with computer engineering
background
Procedure:
1. Submit a video URL to the tool
2. Check and analyse the produced verification report
3. Decide about the video veracity
4. Record the results and the time spent on the task
Labels:
True: If a fake/real video is debunked/verified
False: if the debunking or veryfying of a fake/real video fails
Uncertain: there are indicators that create doubts about the
video credibility but there is no concrete evidence proving that the
video is fake or real.
Is Debunked # videos Time (sec)
True 132 208
False 46 272
Uncertain 22 270
~70% of the fake videos
were succesfully debunked
Is Verified # videos
True 140
False 29
Uncertain 11
~80% of the real videos
were succesfully verified
14. Report Results
1. Start Timer – Don’t forget to stop it when the
verification process is completed
2. Select the Verification Label
3. How certain are you for the Verification Label
4. Verification Features
5. Description: a free text description of the
procedure that you followed to verify the video
6. Submit
https://caa.iti.gr/annotation/
15. Thank you for your attention!
https://caa.iti.gr
https://twitter.com/meverteam
Follow us on Twitter:
Media Verification team website
https://mever.iti.gr/web/
Contact us: Olga Papadopoulou - olgapapa@iti.gr
Symeon Papadopoulos – papadop@iti.gr